Taste Sniffing

Taste sniffing is the workflow's decision-sampling module: it captures your preferences and principles at key trade-off points and turns them into reusable memory.

Position in the System

• Trigger layer: decision points appear in task/plan/build/review
• Collection layer: choices are gathered by questioning or contextual understanding
• Recognition layer: taste-recognition outputs structured payloads
• Write layer: lingxi-memory-write governs and persists
• Reuse layer: memory-retrieve reuses memory in pre/post phases

Trigger Principles

Taste sniffing is not “always ask”; it is “ask when a decision matters”:

• solution trade-offs
• convention exceptions
• acceptance boundaries

Before asking, the system first tries memory retrieval.
If existing memory already covers the context, it can be reused directly.

Output Contract

Taste-sniffing output shares the same memory contract as /remember:

• same extended payload structure
• same write governance and gating
• different source marker (workflow-origin is typically source=choice)

Responsibility Split

• taste-recognition: decides whether and what to capture
• lingxi-memory-write: decides how to govern and persist
• memory-retrieve: decides when to apply memory during execution

Documentation Boundary

This page defines responsibilities and trigger principles only. It does not freeze:

• per-phase question templates
• internal rule tables and scoring details
• field-mapping implementation specifics

For those details, use the repository implementation and internal rules.

Next Steps

• Memory System
• How to Recognize Developer Taste
• Commands and Workflow Reference